No description available.

No description available.

Parsing is the task of assigning syntactic or semantic structure to a natural language sentence. This thesis focuses on syntactic parsing with Combinatory Categorial Grammar (CCG; Steedman 2000). CCG allows incremental processing, which is essential for speech recognition and some machine translation models, and it can build semantic structure in tandem with syntactic parsing. Supertagging solves a subset of the parsing task by assigning lexical types to words in a sentence using a sequence model. It has emerged as a way to improve the efficiency of full CCG parsing (Clark and Curran, 2007) by reducing the parser’s search space. This has been very successful and it is the central theme of this thesis. We begin by an analysis of how efficiency is being traded for accuracy in supertagging. Pruning the search space by supertagging is inherently approximate and to contrast this we include A* in our analysis, a classic exact search technique. Interestingly, we find that combining the two methods improves efficiency but we also demonstrate that excessive pruning by a supertagger significantly lowers the upper bound on accuracy of a CCG parser. Inspired by this analysis, we design a single integrated model with both supertagging and parsing features, rather than separating them into distinct models chained together in a pipeline. To overcome the resulting complexity, we experiment with both loopy belief propagation and dual decomposition approaches to inference, the first empirical comparison of these algorithms that we are aware of on a structured natural language processing problem. Finally, we address training the integrated model. We adopt the idea of optimising directly for a task-specific metric such as is common in other areas like statistical machine translation. We demonstrate how a novel dynamic programming algorithm enables us to optimise for F-measure, our task-specific evaluation metric, and experiment with approximations, which prove to be excellent substitutions. Each of the presented methods improves over the state-of-the-art in CCG parsing. Moreover, the improvements are additive, achieving a labelled/unlabelled dependency F-measure on CCGbank of 89.3%/94.0% with gold part-of-speech tags, and 87.2%/92.8% with automatic part-of-speech tags, the best reported results for this task to date. Our techniques are general and we expect them to apply to other parsing problems, including lexicalised tree adjoining grammar and context-free grammar parsing.

Word-Sense Disambiguation (WSD), holds promise for many NLP applications requiring broad-coverage language understanding, such as summarization (Barzilay and Elhadad, 1997) and question answering (Ramakrishnan et al., 2003). Recent studies have also shown that WSD can benefit machine translation (Vickrey et al., 2005) and information retrieval (Stokoe, 2005). Much work has focused on the computational treatment of sense ambiguity, primarily using data-driven methods. The most accurate WSD systems to date are supervised and rely on the availability of sense-labeled training data. This restriction poses a significant barrier to widespread use of WSD in practice, since such data is extremely expensive to acquire for new languages and domains. Unsupervised WSD holds the key to enable such application, as it does not require sense-labeled data. However, unsupervised methods fall far behind supervised ones in terms of accuracy and ease of use. In this thesis we explore the reasons for this, and present solutions to remedy this situation. We hypothesize that one of the main problems with unsupervised WSD is its lack of a standard formulation and general purpose tools common to supervised methods. As a first step, we examine existing approaches to unsupervised WSD, with the aim of detecting independent principles that can be utilized in a general framework. We investigate ways of leveraging the diversity of existing methods, using ensembles, a common tool in the supervised learning framework. This approach allows us to achieve accuracy beyond that of the individual methods, without need for extensive modification of the underlying systems. Our examination of existing unsupervised approaches highlights the importance of using the predominant sense in case of uncertainty, and the effectiveness of statistical similarity methods as a tool for WSD. However, it also serves to emphasize the need for a way to merge and combine learning elements, and the potential of a supervised-style approach to the problem. Relying on existing methods does not take full advantage of the insights gained from the supervised framework. We therefore present an unsupervised WSD system which circumvents the question of actual disambiguation method, which is the main source of discrepancy in unsupervised WSD, and deals directly with the data. Our method uses statistical and semantic similarity measures to produce labeled training data in a completely unsupervised fashion. This allows the training and use of any standard supervised classifier for the actual disambiguation. Classifiers trained with our method significantly outperform those using other methods of data generation, and represent a big step in bridging the accuracy gap between supervised and unsupervised methods. Finally, we address a major drawback of classical unsupervised systems – their reliance on a fixed sense inventory and lexical resources. This dependence represents a substantial setback for unsupervised methods in cases where such resources are unavailable. Unfortunately, these are exactly the areas in which unsupervised methods are most needed. Unsupervised sense-discrimination, which does not share those restrictions, presents a promising solution to the problem. We therefore develop an unsupervised sense discrimination system. We base our system on a well-studied probabilistic generative model, Latent Dirichlet Allocation (Blei et al., 2003), which has many of the advantages of supervised frameworks. The model’s probabilistic nature lends itself to easy combination and extension, and its generative aspect is well suited to linguistic tasks. Our model achieves state-of-the-art performance on the unsupervised sense induction task, while remaining independent of any fixed sense inventory, and thus represents a fully unsupervised, general purpose, WSD tool.

The amount of information available on the web is growing constantly. As a result, theproblem of retrieving any desired information is getting more difficult by the day. Toalleviate this problem, several techniques are currently being used, both for locatingpages of interest and for extracting meaningful information from the retrieved pages.Information extraction (IE) is one such technology that is used for summarizingunrestricted natural language text into a structured set of facts. IE is already being appliedwithin several domains such as news transcripts, insurance information, and weatherreports. Various approaches to IE have been taken and a number of significant resultshave been reported.In this thesis, we describe the application of IE techniques to the domain of universityweb pages. This domain is broader than previously evaluated domains and has a varietyof idiosyncratic problems to address. We present an analysis of the domain of universityweb pages and the consequences of having them input to IE systems. We then presentUniversityIE, a system that can search a web site, extract relevant pages, and processthem for information such as admission requirements or general information. TheUniversityIE system, developed as part of this research, contributes three IE methods anda web-crawling heuristic that worked relatively well and predictably over a test set ofuniversity web sites.We designed UniversityIE as a generic framework for plugging in and executing IEmethods over pages acquired from the web. We also integrated in the system a genericweb crawler (built at the University of Kentucky) and ported to Java and integrated anexternal word lexicon (WordNet) and a syntax parser (Link Grammar Parser).

This thesis focuses on robust analysis of natural language semantics. A primary bottleneck for semantic processing of text lies in the scarcity of high-quality and large amounts of annotated data that provide complete information about the semantic structure of natural language expressions. In this dissertation, we study statistical models tailored to solve problems in computational semantics, with a focus on modeling structure that is not visible in annotated text data. We first investigate supervised methods for modeling two kinds of semantic phenomena in language. First, we focus on the problem of paraphrase identification, which attempts to recognize whether two sentences convey the same meaning. Second, we concentrate on shallow semantic parsing, adopting the theory of frame semantics (Fillmore, 1982). Frame semantics offers deep linguistic analysis that exploits the use of lexical semantic properties and relationships among semantic frames and roles. Unfortunately, the datasets used to train our paraphrase and frame-semantic parsing models are too small to lead to robust performance. Therefore, a common trait in our methods is the hypothesis of hidden structure in the data. To this end, we employ conditional log-linear models over structures, that are firstly capable of incorporating a wide variety of features gathered from the data as well as various lexica, and secondly use latent variables to model missing information in annotated data. Our approaches towards solving these two problems achieve state-of-the-art accuracy on standard corpora. For the frame-semantic parsing problem, we present fast inference techniques for jointly modeling the semantic roles of a given predicate. We experiment with linear program formulations, and use a commercial solver as well as an exact dual decomposition technique that breaks the role labeling problem into several overlapping components. Continuing with the theme of hypothesizing hidden structure in data for modeling natural language semantics, we present methods to leverage large volumes of unlabeled data to improve upon the shallow semantic parsing task. We work within the framework of graph-based semi-supervised learning, a powerful method that associates similar natural language types, and helps propagate supervised annotations to unlabeled data. We use this framework to improve frame-semantic parsing performance on unknown predicates that are absent in annotated data. We also present a family of novel objective functions for graph-based learning that result in sparse probability measures over graph vertices, a desirable property for natural language types. Not only are these objectives easier to numerically optimize, but also they result in smoothed distributions over predicates that are smaller in size. The experiments presented in this dissertation empirically demonstrates that missing information in text corpora contain considerable semantic information that can be incorporated into structured models for semantics, to significant benefit over the current state of the art. The methods in this thesis were originally presented by Das and Smith (2009, 2011, 2012), and Das et al. (2010, 2012). The thesis gives a more thorough exposition, relating and comparing the methods, and also presents several extensions of the aforementioned papers.

Determining the quality of natural language applications is one of the most important aspects of technology development. There has, however, been very little work done on establishing how well the methods and measures represent the quality of the technology and how reliable the evaluation results presented in most research are. This study presents a new stepwise evaluation reliability methodology that provides a step-by-step framework for creating predictive models of evaluation metric reliability that take into account inherent evaluation variables. These models can then be used to predict how reliable a particular evaluation will be prior to doing an evaluation, based on the variables that are present in the evaluation data. This allows evaluators to predict the reliability of the evaluation prior to doing the evaluation and adjusting the evaluation data to ensure reliable results. Furthermore, this permits researchers to compare results when the same evaluation data is not available. The new methodology is firstly applied to a well-defined technology, namely spelling checkers, with a detailed discussion of the evaluation techniques and statistical procedures required to accurately model an evaluation. The spelling checker evaluations are investigated in more detail to show how individual variables affect the evaluation results. Finally, a predictive regression model for each of the spelling checker evaluations is created and validated to verify the accuracy of its predictive capability. After performing the in-depth analysis and application of the stepwise evaluation reliability methodology on spelling checkers, the methodology is applied to two more technologies, namely part of speech tagging and named entity recognition. These validation procedures are applied across multiple languages, specifically Dutch, English, Spanish and Iberian Portuguese. Performing these additional evaluations shows that the methodology is applicable to a broader set of technologies across multiple languages. / Thesis (PhD (Linguistics and Literary Theory))--North-West University, Potchefstroom Campus, 2013.

Determining the quality of natural language applications is one of the most important aspects of technology development. There has, however, been very little work done on establishing how well the methods and measures represent the quality of the technology and how reliable the evaluation results presented in most research are. This study presents a new stepwise evaluation reliability methodology that provides a step-by-step framework for creating predictive models of evaluation metric reliability that take into account inherent evaluation variables. These models can then be used to predict how reliable a particular evaluation will be prior to doing an evaluation, based on the variables that are present in the evaluation data. This allows evaluators to predict the reliability of the evaluation prior to doing the evaluation and adjusting the evaluation data to ensure reliable results. Furthermore, this permits researchers to compare results when the same evaluation data is not available. The new methodology is firstly applied to a well-defined technology, namely spelling checkers, with a detailed discussion of the evaluation techniques and statistical procedures required to accurately model an evaluation. The spelling checker evaluations are investigated in more detail to show how individual variables affect the evaluation results. Finally, a predictive regression model for each of the spelling checker evaluations is created and validated to verify the accuracy of its predictive capability. After performing the in-depth analysis and application of the stepwise evaluation reliability methodology on spelling checkers, the methodology is applied to two more technologies, namely part of speech tagging and named entity recognition. These validation procedures are applied across multiple languages, specifically Dutch, English, Spanish and Iberian Portuguese. Performing these additional evaluations shows that the methodology is applicable to a broader set of technologies across multiple languages. / Thesis (PhD (Linguistics and Literary Theory))--North-West University, Potchefstroom Campus, 2013.

Thesauri are important tools for many Natural Language Processing applications. Roget's Thesaurus is particularly useful. It is of high quality and has been in development for over a century and a half. Yet its applications have been limited, largely because the only publicly available edition dates from 1911. This thesis proposes and tests methods of automatically updating the vocabulary of the 1911 Roget’s Thesaurus. I use the Thesaurus as a source of training data in order to learn from Roget’s for the purpose of updating Roget’s. The lexicon is updated in two stages. First, I develop a measure of semantic relatedness that enhances existing distributional techniques. I improve existing methods by using known sets of synonyms from Roget’s to train a distributional measure to better identify near synonyms. Second, I use the new measure of semantic relatedness to find where in Roget’s to place a new word. Existing words from Roget’s are used as training data to tune the parameters of three methods of inserting words. Over 5000 new words and word-senses were added using this process. I conduct two kinds of evaluation on the updated Thesaurus. One is on the procedure for updating Roget’s. This is accomplished by removing some words from the Thesaurus and testing my system's ability to reinsert them in the correct location. Human evaluation of the newly added words is also performed. Annotators must determine whether a newly added word is in the correct location. They found that in most cases the new words were almost indistinguishable from those already existing in Roget's Thesaurus. The second kind of evaluation is to establish the usefulness of the updated Roget’s Thesaurus on actual Natural Language Processing applications. These applications include determining semantic relatedness between word pairs or sentence pairs, identifying the best synonym from a set of candidates, solving SAT-style analogy problems, pseudo-word-sense disambiguation, and sentence ranking for text summarization. The updated Thesaurus consistently performed at least as well or better the original Thesaurus on all these applications.

This thesis investigates the evaluation of automatic and manual text segmentation. Text segmentation is the process of placing boundaries within text to create segments according to some task-dependent criterion. An example of text segmentation is topical segmentation, which aims to segment a text according to the subjective definition of what constitutes a topic. A number of automatic segmenters have been created to perform this task, and the question that this thesis answers is how to select the best automatic segmenter for such a task. This requires choosing an appropriate segmentation evaluation metric, confirming the reliability of a manual solution, and then finally employing an evaluation methodology that can select the automatic segmenter that best approximates human performance. A variety of comparison methods and metrics exist for comparing segmentations (e.g., WindowDiff, Pk), and all save a few are able to award partial credit for nearly missing a boundary. Those comparison methods that can award partial credit unfortunately lack consistency, symmetricity, intuition, and a host of other desirable qualities. This work proposes a new comparison method named boundary similarity (B) which is based upon a new minimal boundary edit distance to compare two segmentations. Near misses are frequent, even among manual segmenters (as is exemplified by the low inter-coder agreement reported by many segmentation studies). This work adapts some inter-coder agreement coefficients to award partial credit for near misses using the new metric proposed herein, B. The methodologies employed by many works introducing automatic segmenters evaluate them simply in terms of a comparison of their output to one manual segmentation of a text, and often only by presenting nothing other than a series of mean performance values (along with no standard deviation, standard error, or little if any statistical hypothesis testing). This work asserts that one segmentation of a text cannot constitute a “true” segmentation; specifically, one manual segmentation is simply one sample of the population of all possible segmentations of a text and of that subset of desirable segmentations. This work further asserts that an adapted inter-coder agreement statistics proposed herein should be used to determine the reproducibility and reliability of a coding scheme and set of manual codings, and then statistical hypothesis testing using the specific comparison methods and methodologies demonstrated herein should be used to select the best automatic segmenter. This work proposes new segmentation evaluation metrics, adapted inter-coder agreement coefficients, and methodologies. Most important, this work experimentally compares the state-or-the-art comparison methods to those proposed herein upon artificial data that simulates a variety of scenarios and chooses the best one (B). The ability of adapted inter-coder agreement coefficients, based upon B, to discern between various levels of agreement in artificial and natural data sets is then demonstrated. Finally, a contextual evaluation of three automatic segmenters is performed using the state-of-the art comparison methods and B using the methodology proposed herein to demonstrate the benefits and versatility of B as opposed to its counterparts.